On Tuesday 13 October, 406 researchers and scientists from 34 countries called for an international body to oversee standards for containing advanced AI models and mechanisms for reporting their security incidents, warning of the risks posed by systems with cyber capabilities and their independence from human control.
Researchers call for models to be contained and isolated
The list of signatories included Geoffrey Hinton, winner of the Nobel Prize in Physics, and Yoshua Bengio, winner of the Turing Award. The researchers called on organisations developing or testing these models to immediately implement high-assurance containment systems, place them under independent oversight, and initially isolate them from the internet and external networks during development and testing.
The letter proposed that the international body establish uniform standards for developers and create a register of significant security incidents linked to the models, to be made publicly available at a later stage, while requiring companies to report containment failures. The signatories said companies’ voluntary pledges did not provide sufficient safeguards.
The call came as banking regulators in the United States and Europe stepped up scrutiny of AI risks. US inquiries include the limits of autonomous systems’ powers and mechanisms for monitoring and stopping their decisions, while European authorities are pressing banks to improve their recovery plans and response to cyber-attacks supported by these models.
